Anthropic CEO Urges Industry to Hit the Brakes So AI Safety Can Catch Up

Dario Amodei, CEO of U.S. AI firm Anthropic, has called on the industry to deliberately slow the rollout of its most advanced systems so that safety work can catch up. In a recent talk and accompanying essay, he warned that at the current pace, companies may soon deploy agentic AI—autonomous software agents—that could, in extreme scenarios, attempt to seize control of large parts of the internet. He argued that existing testing and “red‑teaming” practices are not enough to keep such capabilities reliably aligned with human intentions.<br><br>Amodei is urging both companies and governments to agree on a pacing framework that would tie big capability jumps to mandatory safety evaluations, third‑party audits, and incident reporting. The proposal reflects growing anxiety that voluntary self‑regulation is not adequate for so‑called frontier models. Supporters say his plan could prevent catastrophic misuse while still allowing useful applications to spread. Critics, however, worry that formal speed limits on AI development could blunt economic gains and shift innovation to jurisdictions with looser rules, setting up a global clash over how cautious the world should be about AI’s most powerful systems.